• Distant Viewing: Computational Image Similarity and Visual Resources Collections

    Author(s):
    Tess Colwell (see profile) , Lindsay King
    Date:
    2022
    Subject(s):
    Digital humanities
    Item Type:
    Conference paper
    Conf. Title:
    Art Libraries Society of North America Annual Conference, 2022
    Permanent URL:
    https://doi.org/10.17613/qs39-ch03
    Abstract:
    What is the role of a legacy visual resources collection in an era of ubiquitous digital images? In what ways can we think about images as data? In this presentation, the project team will demonstrate a new tool that algorithmically groups digital images into clusters based solely on visually similar characteristics, rather than traditional categories like chronology, creator, subject, geography, etc. Metadata can aid in further analysis and curation, but is not a factor in the initial clustering, which relies on a convolutional neural network. Our dataset is more than 370,000 digital images scanned from lantern slides, 35mm slides, and photographs, related to global art, architecture, and material culture to support teaching and research in the arts and humanities. Descriptive metadata ranges from comprehensive and detailed item-level information to minimal captions. Running randomly-sampled subsets of the whole dataset through the visual similarity tool has revealed previously-known areas of focus in addition to unusual juxtapositions and observations about how the algorithm works. The session will include context and background on the project, a demonstration of the tool and new curation features, and a discussion on next steps and how these features allow us to analyze the collection in unprecedented ways. What does this kind of visualization, the digital equivalent of moving slides around on a giant light table, tell us about a collection of images accumulated for teaching and research over more than six decades by an art history department? How can this kind of distant viewing make the images more meaningful to students and faculty? 
    Metadata:
    Status:
    Published
    Last Updated:
    1 year ago
    License:
    Attribution-NonCommercial

    Downloads

    Item Name: pdf arlis2022_distant-viewing_-computational-image-similarity-and-visual-resources-collections.pdf
      Download View in browser
    Activity: Downloads: 18